Rydberg Electrometry

A Rydberg electrometer infers an electric-field property from the large dipole response of a highly excited atom. In the most common vapor-cell implementation, two optical fields create Rydberg electromagnetically induced transparency (EIT), a radio-frequency field couples the addressed Rydberg state to a neighboring state, and probe transmission supplies the data. The familiar Autler–Townes formula can then convert a resolved spectral splitting into a local field amplitude.

That one-line conversion is the middle of a measurement, not its beginning or end. A complete result must say which field component and amplitude convention are estimated, how a laser scan maps onto an atomic energy interval, how the glass cell changes the incident field, which atoms and positions contribute, and how the recorded photons determine an uncertainty. These distinctions are especially important when the splitting is unresolved, the field is detuned, or the desired measurand is the field outside the cell.

Required background. Rydberg atoms provides transition dipoles and level spacings; EIT explains the optical dark resonance; Autler–Townes splitting distinguishes dressed gaps from observed peaks. Parameter estimation supplies the likelihood and uncertainty calculation.

Electric-field amplitudes and optical readout

For a nearly monochromatic field, write the real field as

$$\mathbf E(\mathbf r,t) = \operatorname{Re} \left[ \mathbf E_0(\mathbf r) e^{-i\omega t} \right].$$

$\mathbf E_0$ is a peak complex phasor. For a linearly polarized sinusoid, $E_{\mathrm{rms}}=E_0/\sqrt 2$. Autler–Townes Rabi frequencies below use the peak convention. For elliptical polarization, define the phasor norm by $E_0=\sqrt{\mathbf E_0^*\cdot\mathbf E_0}$. It equals $\sqrt{2\langle|\mathbf E(t)|^2\rangle_t}$, but need not equal the maximum instantaneous vector magnitude. This distinction matters when comparing linear and circular polarization.

Four-Level RF-to-Optical Transducer

States, fields, and detunings

Use a ladder $|g\rangle\leftrightarrow|e\rangle\leftrightarrow|r\rangle$ with probe and coupling Rabi coefficients $\Omega_p$ and $\Omega_c$. A field near angular frequency $\omega_\mu$ couples $|r\rangle$ to a higher Rydberg state $|s\rangle$ with coefficient $\Omega_\mu$. The displayed ladder assumes $E_s>E_r$; a downward RF transition requires the corresponding rotating-frame and phasor-conjugation changes. Define atom-minus-field detunings

$$\begin{aligned} \Delta_p &= \omega_{eg}-\omega_p,\\ \Delta_c &= \omega_{re}-\omega_c,\\ \Delta_\mu &= \omega_{sr}-\omega_\mu, \end{aligned}$$

and cumulative detunings

$$\delta_2 = \Delta_p+\Delta_c, \qquad \delta_3 = \Delta_p+\Delta_c+\Delta_\mu.$$

Angular frequencies are used unless $2\pi$ is written explicitly. In a rotating-wave frame, one convenient Hamiltonian is

$$\frac{H}{\hbar} = \begin{pmatrix} 0 & \Omega_p^*/2 & 0 & 0\\ \Omega_p/2 & \Delta_p & \Omega_c^*/2 & 0\\ 0 & \Omega_c/2 & \delta_2 & \Omega_\mu^*/2\\ 0 & 0 & \Omega_\mu/2 & \delta_3 \end{pmatrix}.$$

The model is useful only after specifying the states hidden inside each label. Hyperfine and Zeeman sublevels, unwanted polarization components, nearby Rydberg manifolds, and off-resonant couplings can turn one four-state diagram into many interfering pathways.

The projected RF coupling

For peak phasor $\mathbf E_0$ and a selected transition,

$$\Omega_\mu = -\frac{ \langle s|\mathbf d|r\rangle \mathbin{\cdot} \mathbf E_0 }{\hbar}.$$

Its magnitude is

$$|\Omega_\mu| = \frac{d_{\mathrm{eff}}E_0}{\hbar}, \qquad d_{\mathrm{eff}} = \left| \langle s|\mathbf d|r\rangle \mathbin{\cdot} \widehat{\boldsymbol\epsilon} \right|.$$

The sign and complex phase matter in a closed-loop or coherent-mixing experiment, even though an isolated splitting depends only on $|\Omega_\mu|$. The rotating-wave approximation also requires the counter-rotating and neighboring-transition corrections to remain small on the required uncertainty scale.

Weak-probe optical response

For Markovian coherence damping in the weak-probe limit, $\rho_{gg}\simeq1$. Solving the three linear equations for $\rho_{ge}$, $\rho_{gr}$, and $\rho_{gs}$ gives a continued fraction. With coherence half-widths $\gamma_{ge}$, $\gamma_{gr}$, and $\gamma_{gs}$,

$$\frac{2\rho_{ge}}{\Omega_p^*} = \frac{i}{ \gamma_{ge}-i\Delta_p + \dfrac{|\Omega_c|^2/4}{ \gamma_{gr}-i\delta_2 + \dfrac{|\Omega_\mu|^2/4}{ \gamma_{gs}-i\delta_3 } } }.$$

This expression displays the nested transduction: the RF field changes a Rydberg coherence, the coupling field maps that change onto the optical coherence, and propagation maps the coherence onto transmitted light. It is not a universal vapor-cell fit function. Saturation, optical pumping, additional levels, velocity changes during transit, and propagation through appreciable optical depth require optical Bloch or master-equation modeling.

For a scalar probe polarization with peak phasor $E_p$, use $\Omega_p=-d_{eg}E_p/\hbar$ and positive-frequency dependence $e^{-i\omega_p t}$. If $n_a$ is number density, the physical susceptibility is

$$\chi=-K\frac{\rho_{eg}}{\Omega_p} =-K\left(\frac{\rho_{ge}}{\Omega_p^*}\right)^*, \qquad K=\frac{2n_a|d_{eg}|^2}{\epsilon_0\hbar}>0.$$

The minus sign comes from $-\mathbf d\cdot\mathbf E$ and the conjugation from the positive-frequency polarization. Consequently $\operatorname{Im}\chi=K\operatorname{Im}(\rho_{ge}/\Omega_p^*)$, whereas the real parts have opposite signs. Confusing the coherence with $\chi$ preserves this absorption proxy but reverses dispersion.

For a dilute homogeneous slice with $|\chi|\ll1$, negligible reflections, and a spatially uniform unsaturated response, the intensity transmission is

$$T_p(\Delta_p) = \exp\!\left[ -k_p L\operatorname{Im}\chi(\Delta_p) \right].$$

If spectral bin $j$ records a photon count $n_j$, a basic detector model is

$$n_j \sim \operatorname{Poisson} \left[ \mu_j(\theta,\boldsymbol\lambda) \right],$$

where $\boldsymbol\lambda$ collects laser power, detunings, linewidths, background, gain, and other nuisance parameters. Fitting peak positions is one possible estimator. A correctly specified full-line-shape likelihood can retain information lost by peak picking, especially near the resolution boundary; a more detailed but incorrect model need not reduce bias.

Response calculated from the continued fraction, not from fitted Gaussian peaks. The probe and RF are resonant, while the coupling laser is scanned. In units of $\gamma_{ge}$, take $\Omega_c=1.5$, $\gamma_{gr}=\gamma_{gs}=0.03$, and $k_pLK/(2\gamma_{ge})=2$. The RF-off transparency peak becomes a narrow central dip and then two separated transmission peaks as $|\Omega_\mu|$ increases. The single-velocity, uniform-field model illustrates why a spectral change can be measurable before two peak positions give a reliable field estimate. Thermal and spatial averaging are omitted.

Resolved Autler–Townes Electrometry

Resonant and detuned dressed gaps

Ignore the weak optical probe momentarily and diagonalize the RF-coupled Rydberg pair. Relative to $|r\rangle$, its rotating-frame block can be written

$$\frac{H_{rs}}{\hbar} = \begin{pmatrix} 0 & \Omega_\mu^*/2\\ \Omega_\mu/2 & \Delta_\mu \end{pmatrix}.$$

The dressed shifts are

$$\lambda_\pm = \frac{\Delta_\mu}{2} \pm \frac12 \sqrt{ \Delta_\mu^2+|\Omega_\mu|^2 },$$

so the eigenvalue gap is

$$S_\omega = \sqrt{ \Delta_\mu^2+|\Omega_\mu|^2 }.$$

On resonance, $S_\omega=|\Omega_\mu|$. If a justified spectral estimator returns the corresponding splitting $S_\nu=S_\omega/(2\pi)$ in hertz, then

$$E_{0,\mathrm{atom}} = \frac{hS_\nu}{d_{\mathrm{eff}}}.$$

This direct conversion applies to the local, projected, peak field when the RF is resonant and the observed peak separation is a valid estimator of the dressed gap. Off resonance,

$$E_{0,\mathrm{atom}} = \frac{\hbar}{d_{\mathrm{eff}}} \sqrt{S_\omega^2-\Delta_\mu^2}.$$

Near the resolution threshold, raw local maxima need not coincide with the real parts of the response poles. The full line shape should then determine $\Omega_\mu$ and $\Delta_\mu$ jointly.

The scan-axis correction

The atomic gap and the separation on a plotted laser-frequency axis are not always equal. Let a moving atom have signed axial wave vectors $k_p$ and $k_c$. Its optical detunings are

$$\Delta_p(v) = \Delta_p+k_pv, \qquad \Delta_c(v) = \Delta_c+k_cv.$$

When the coupling laser is scanned while a resonant probe selects atoms near $v=0$, the coupling-axis separation directly gives

$$\Delta\nu_c = \frac{|\Omega_\mu|}{2\pi}$$

in the ideal resonant limit. If instead the probe is scanned and the coupling frequency is held fixed, the strongly absorbing velocity class approximately satisfies $\Delta_p(v)=0$. For counter-propagating beams this gives

$$\delta_2 \simeq -\frac{k_c}{k_p}\Delta_p,$$

and therefore

$$\Delta\nu_p \simeq \frac{\nu_p}{\nu_c} \frac{|\Omega_\mu|}{2\pi} = \frac{\lambda_c}{\lambda_p} \frac{|\Omega_\mu|}{2\pi}.$$

Thus a probe-axis splitting must be divided by $\nu_p/\nu_c=\lambda_c/\lambda_p$ before applying the atomic conversion. This result follows from the stated scan and propagation conventions; a memorized wavelength factor can be inverted by changing either one. Finite linewidth, optical pumping, temperature, and multilevel structure can also modify the simple selected-velocity argument, so precision work should fit the velocity-averaged response.

Below the Resolved-Splitting Regime

Line-shape estimation

A field does not become undetectable merely because two peaks cannot be resolved. It can change the EIT depth, width, curvature, or phase. For independent Poisson spectral bins, the classical Fisher information for a field parameter $E$ is

$$F_E = \sum_j \frac{1}{\mu_j(E)} \left( \frac{\partial\mu_j(E)}{\partial E} \right)^2.$$

With nuisance parameters $\boldsymbol\lambda$, an invertible full Fisher matrix gives the local bound for an estimator locally unbiased in the field and nuisance directions,

$$\operatorname{Var}(\widehat E) \geq \left[F^{-1}\right]_{EE},$$

not from $1/F_E$ while pretending that laser detuning and contrast are known. The best probe detuning is usually where the expected signal changes steeply and reproducibly, not necessarily at line center. Modulation and balanced or homodyne optical readout can move the estimate away from low-frequency laser noise, but their duty cycle and reference noise belong in the comparison.

If the observable is even in an unknown field, its small-field response may begin as

$$y(E) = y(0)+aE^2+O(E^4).$$

Then $\partial_Ey=0$ at $E=0$. In a regular likelihood with positive noise floor, field-amplitude Fisher information vanishes there, although $E^2$ can still be estimable. A signed field and its negative are indistinguishable in this response; a nonnegative amplitude at zero is a parameter boundary, where the usual unbiased Cramér–Rao reasoning needs care. A known bias or coherent local oscillator restores a linear term.

Atomic homodyne and heterodyne mixing

Let a strong local oscillator and a weak signal address the same effective Rydberg transition:

$$\begin{aligned} E_{\mathrm{LO}}(t) &= E_L\cos(\omega_Lt),\\ E_{\mathrm s}(t) &= E_s\cos(\omega_st+\phi). \end{aligned}$$

Assume the LO and signal have the same polarization projection and spatial mode. For $E_s\ll E_L$ and $\omega_s-\omega_L=\omega_{\mathrm{IF}}$, the field-envelope magnitude is

$$E_{\mathrm{env}}(t) \simeq E_L + E_s\cos(\omega_{\mathrm{IF}}t+\phi).$$

If the modulation is slow enough for a quasistatic optical response $y=f(E_{\mathrm{env}})$, linearization gives

$$y(t) \simeq f(E_L) + f'(E_L)E_s \cos(\omega_{\mathrm{IF}}t+\phi).$$

The atoms now down-convert the signal to an intermediate-frequency optical beat note. At nonzero IF, two calibrated demodulation phases estimate $E_s\cos\phi$ and $E_s\sin\phi$; the beat amplitude estimates $E_s$. At zero IF, one fixed homodyne phase measures only one quadrature. Near the response bandwidth, replace the real static slope by a calibrated complex frequency response, including its attenuation and phase lag. This extends detection below the resolved Autler–Townes regime and enables coherent reception. The result is not reference-free: local-oscillator amplitude, phase noise, spatial overlap, and detuning are resources and nuisance parameters.

Resonant, Off-Resonant, and Static Fields

Rydberg electrometry spans several physical responses. They should not be collapsed into one calibration formula.

Regime	Dominant observable	Leading field dependence	Information lost without a reference
resonant RF, resolved	Autler–Townes gap	$S_\nu\propto E_0$	absolute carrier phase
resonant RF, unresolved	EIT line shape or mixer beat	model dependent; linear with a coherent bias	phase or sign in an intensity-only measurement
far-detuned RF	differential AC Stark shift	$\Delta\nu\propto E_0^2$	sign and generally carrier phase
static or quasistatic	differential DC Stark shift	$\Delta\nu\propto E_{\mathrm{dc}}^2$ absent a bias	field sign

For a nondegenerate state with scalar static polarizability $\alpha_i(0)$,

$$\Delta U_i^{(\mathrm{dc})} = -\frac12\alpha_i(0)E_{\mathrm{dc}}^2.$$

The measured transition shift is set by the differential polarizability,

$$\Delta\nu_{sr}^{(\mathrm{dc})} = -\frac{ \alpha_s(0)-\alpha_r(0) }{2h} E_{\mathrm{dc}}^2.$$

For a far-off-resonant sinusoid with peak amplitude $E_0$, the cycle average introduces another factor of $1/2$:

$$\Delta U_i^{(\mathrm{ac})} = -\frac14 \alpha_i(\omega)E_0^2.$$

Tensor polarizability, near-degenerate mixing, and strong fields require a Stark-map or Floquet calculation rather than these scalar formulas. The canonical derivations are on Stark Effect in Atoms and Dynamic Polarizability.

A known static bias $E_b$ can linearize a small perturbation $e$:

$$(E_b+e)^2 = E_b^2+2E_be+e^2.$$

It also imports bias calibration and drift. In glass alkali-vapor cells, surface charge can screen low-frequency external fields or create inhomogeneous offsets, so the field at the atoms must be measured or modeled rather than inferred from electrode voltage alone.

Polarization and Vector Reconstruction

Choose a quantization axis and spherical basis $\mathbf e_0=\mathbf e_z$, $\mathbf e_{\pm1}=\mp(\mathbf e_x\pm i\mathbf e_y)/\sqrt2$. Expand the RF phasor using coefficients

$$E_q = \mathbf e_q^* \mathbin{\cdot} \mathbf E_0, \qquad q\in\{-1,0,+1\}.$$

Define the standard tensor operators $d_0=d_z$, $d_{\pm1}=\mp(d_x\pm i d_y)/\sqrt2$. With these definitions, $\mathbf d\cdot\mathbf E_0=\sum_q d_qE_q$: the field expansion coefficients use the conjugated basis, whereas the tensor operators do not. For one angular-momentum pathway,

$$\Omega_{m m'}^{(q)} = -\frac{E_q}{\hbar} \langle F'm'|d_q|Fm\rangle,$$

with

$$\langle F'm'|d_q|Fm\rangle = (-1)^{F'-m'} \begin{pmatrix} F'&1&F\\ -m'&q&m \end{pmatrix} \langle F'\|d\|F\rangle.$$

Selection rules therefore make the spectrum a polarization-dependent sum of pathways, not a direct meter of $|\mathbf E|$. Optical pumping changes their weights. Some populated states may be uncoupled and leave an unsplit central feature; others have different dipole moments and broaden or resolve the doublet.

A vector measurement needs enough independent settings to identify the desired phasor parameters. Schematically,

$$\mathbf y = \mathbf f \left( \mathbf E_0, \mathbf B_0, \boldsymbol\epsilon_p, \boldsymbol\epsilon_c, \boldsymbol\lambda \right).$$

Rotating a known quantization axis, changing optical polarizations, resolving Zeeman components, or using multiple transitions can provide independent constraints. The Jacobian with respect to the estimated field components must identify the desired field directions after accounting for nuisance parameters. Intensity-only spectra are unchanged by a common phasor phase; absolute phase requires a coherent reference. Local rank also does not eliminate discrete, globally indistinguishable solutions. Measuring three peak separations is not automatically a three-dimensional vector measurement if the same unknown polarization impurity affects all three.

Thermal Motion and Spatial Transfer

Velocity averaging

For velocity distribution $f(v)$, the susceptibility of a dilute thermal ensemble begins with

$$\overline\chi(\Delta_p,\Delta_c) = \int_{-\infty}^{\infty} f(v) \chi \left[ \Delta_p+k_pv, \Delta_c+k_cv \right] \,dv.$$

Counter-propagating probe and coupling beams reduce the residual two-photon Doppler wave vector $k_p+k_c$, but unequal wavelengths prevent perfect cancellation in the common two-photon ladder. Beam angle, transit time, collisions, laser linewidth, and velocity-changing processes further modify the response. Three-photon ladders can be arranged with a smaller net wave vector, trading additional lasers and levels for narrower Doppler response.

The field inside a vapor cell

At carrier frequency $\omega$, write the local field as

$$\mathbf E_{\mathrm{atom}}(\mathbf r,\omega) = \mathsf T_{\mathrm{cell}}(\mathbf r,\omega) \mathbf E_{\mathrm{external}}(\omega).$$

$\mathsf T_{\mathrm{cell}}$ can include dielectric boundary conditions, standing waves, nearby conductors, electrodes, supports, resonators, and polarization rotation. At low frequency it can include charge transport and screening. It is generally a complex, position-dependent tensor rather than a single scalar correction. This linear map assumes fixed material response and geometry; field-dependent surface charge or nonlinear response requires a more general forward model.

The detected spectrum is weighted by probe and coupling intensities and by atomic trajectories. A schematic forward model is

$$\mu_j = \mathcal D_j \left\{ \int d^3r\,dv\, W(\mathbf r,v) \mathcal P \left[ \chi_j (\mathbf E_{\mathrm{atom}},v,\mathbf r) \right] \right\},$$

where the notation only summarizes the operations: the susceptibility must first be averaged over the relevant local velocity distribution, then propagated along each optical path, before detector pixels or modes are combined. These operations generally cannot be interchanged. Nor can an inhomogeneous RF field usually be replaced by its spatial mean. A field gradient may instead produce broadening, unequal peaks, or several peaks.

A fitted local amplitude is therefore an effective parameter of the chosen spatial model, not automatically the arithmetic mean field. An external-field estimate additionally needs a sufficiently identifiable electromagnetic transfer model. A single projected local measurement cannot invert an arbitrary three-component transfer tensor.

A subwavelength optical spot can localize the readout more finely than the RF wavelength, but spatial resolution is set by the full weighting kernel, including beam waist, depth of field, atomic motion, camera pixels, and inverse regularization. It is not simply the cell size.

Sensitivity, Bandwidth, and Dynamic Range

Averaging-time sensitivity

For total averaging time $\tau$ in a stationary white-noise regime, define

$$\sigma_E(\tau) \simeq \frac{\eta_E}{\sqrt\tau},$$

where $\eta_E$ has units $\mathrm{V\,m^{-1}\,Hz^{-1/2}}$ and includes cycle overhead when $\tau$ is elapsed time. This coefficient is defined by estimator averaging, not by a one-sided PSD. A continuous white-noise record with rectangular averaging has one-sided amplitude spectral density $\sqrt2\,\eta_E$. Comparisons need the same local or external field, peak or RMS convention, carrier and modulation frequencies, spatial mode, and observation time.

For a resolved resonant splitting,

$$\sigma_E = \frac{h}{d_{\mathrm{eff}}} \sigma_{S_\nu},$$

before scan-axis, detuning, and transfer uncertainties. This relation converts a splitting-estimator uncertainty; it does not predict that uncertainty. Photon shot noise, laser amplitude and frequency noise, atomic projection or number fluctuations, thermal radiation, detector noise, and model mismatch all enter through the data likelihood.

Three different bandwidths

The word bandwidth has at least three meanings here.

1. Carrier coverage is the set of RF frequencies addressable by available Rydberg transitions or by Stark, Zeeman, or dressing-based tuning. Broad coverage across configurations is not simultaneous broadband reception.
2. Instantaneous modulation bandwidth is the highest envelope or phase variation faithfully followed at one carrier setting. It depends on optical pumping, Rabi frequencies, coherence, transit replacement, and detector electronics.
3. Measurement bandwidth or resolution bandwidth is the noise-equivalent bandwidth used to turn fluctuations into a spectral sensitivity.

Using $e^{-i\omega_m t}$ for modulation, a first-order response with time constant $\tau_R$ gives

$$H(\omega_m) = \frac{H(0)}{1-i\omega_m\tau_R}, \qquad f_{3\mathrm{dB}} = \frac{1}{2\pi\tau_R}.$$

In a thermal vapor, the transit estimate

$$\tau_{\mathrm{tr}} \sim \frac{w}{v_\perp}$$

can be comparable to or shorter than internal relaxation times. Smaller beams can increase response speed but reduce atom number and interaction length, so bandwidth and sensitivity are jointly engineered.

Lower and upper field limits

At low amplitude, an Autler–Townes estimator fails when the line shape no longer identifies two branches. Line-shape, modulation, or mixer estimators can continue below that point. At high amplitude, the selected pair is no longer isolated: nearby Rydberg levels, multiple RF photons, spatially varying splittings, counter-rotating terms, ionization, and detector scan range can limit the model. A trustworthy dynamic range is the interval over which the chosen response model, identifiability, and uncertainty coverage remain valid. Useful calibrated responses can be nonlinear.

Traceability and Uncertainty

The resonant local-field chain is attractive because $h$ is exact in the SI and a transition dipole can be calculated from atomic structure:

$$S_\nu \longrightarrow |\Omega_\mu| \longrightarrow E_{0,\mathrm{atom}}.$$

It is better described as atom-referenced than as automatically self-calibrating. The transition identity, polarization projection, scan-axis scale, RF detuning, and line-shape estimator must still be established. For an external field, one must also fit or invert the cell-transfer model on the field subspace actually identified by the measurements.

For a scalar transfer magnitude $T$ and scan factor $D=S_{\mathrm{scan}}/S_\nu$, the resonant conversion is $E_{\mathrm{external}}=hS_{\mathrm{scan}}/(D d_{\mathrm{eff}}T)$. Here $D=1$ for the ideal coupling scan and $D=\lambda_c/\lambda_p$ for the ideal counter-propagating probe scan. With independent small uncertainties, a first-order estimate is

$$\left( \frac{u(E_{\mathrm{external}})}{E_{\mathrm{external}}} \right)^2 \simeq \left(\frac{u(S_{\mathrm{scan}})}{S_{\mathrm{scan}}}\right)^2 + \left(\frac{u(D)}{D}\right)^2 + \left(\frac{u(d_{\mathrm{eff}})}{d_{\mathrm{eff}}}\right)^2 + \left(\frac{u(T)}{T}\right)^2 +u_{\mathrm{model}}^2.$$

$u_{\mathrm{model}}$ denotes a dimensionless relative model uncertainty. Correlations require covariance terms, and line-shape model discrepancy is often not well represented by a single Gaussian contribution. A robust analysis varies beam position and power, reverses or rotates polarizations, changes scan direction, fits alternative plausible models, maps the cell electromagnetically, and reports which corrections were measured rather than assumed.

Precision and accuracy answer different questions. Repeated spectra can give a small $\sigma_E$ while an incorrect cell transfer, pathway dipole, or frequency scale biases every result. Conversely, a traceable calibration does not guarantee high sensitivity or communications-ready bandwidth.

Worked Calibration Example

Suppose a coupling-laser scan resolves a resonant splitting

$$S_\nu = 1.000\ \mathrm{MHz}$$

for an isolated pathway with

$$d_{\mathrm{eff}} = 1000\,e a_0 = 8.4784\times10^{-27}\ \mathrm{C\,m}.$$

The local peak field is

$$\begin{aligned} E_{0,\mathrm{atom}} &= \frac{hS_\nu}{d_{\mathrm{eff}}}\\ &= 7.815\times10^{-2}\ \mathrm{V\,m^{-1}}\\ &= 0.7815\ \mathrm{mV\,cm^{-1}}. \end{aligned}$$

For a linearly polarized sinusoid,

$$E_{\mathrm{rms,atom}} = 5.526\times10^{-2}\ \mathrm{V\,m^{-1}}.$$

If the same atomic gap were read by scanning a $780\ \mathrm{nm}$ probe while a counter-propagating $480\ \mathrm{nm}$ coupling laser remained fixed, the ideal probe-axis separation would instead be

$$\Delta\nu_p = \frac{480}{780} (1.000\ \mathrm{MHz}) \simeq 0.615\ \mathrm{MHz}.$$

Applying $h\Delta\nu_p/d_{\mathrm{eff}}$ directly would underestimate the field. Finally, if a validated scalar cell model gives $T=0.80$, the inferred external peak field is $E_{0,\mathrm{external}}=0.0977\ \mathrm{V\,m^{-1}}$. The local atomic result does not require that last correction; the external result does.

Choosing an Operating Mode

Goal	Useful mode	Main calibration object	Dominant failure mode
absolute local RF amplitude	resonant resolved splitting	$d_{\mathrm{eff}}$, detuning, scan axis	multilevel or inhomogeneous peaks
very weak coherent signal	RF homodyne or heterodyne	LO amplitude, phase, transfer gain	reference noise and drift
polarization	resolved sublevels and varied axes	angular pathways and optical polarization	rank-deficient reconstruction
off-resonant carrier	AC Stark or dressed-state model	dynamic polarizability or Floquet spectrum	quadratic ambiguity and nearby levels
low-frequency or DC field	DC Stark shift with bias or electrodes	differential polarizability and screening	wall charge and field inhomogeneity
field image	camera or scanned optical volume	spatial kernel and cell map	interpreting pixel size as resolution
communication waveform	locked optical readout and mixer	transfer function and linear range	confusing data rate with carrier coverage

No mode is universally best. Resolved splitting provides a transparent frequency-to-field conversion but has a linewidth-imposed lower boundary. Mixing provides weak-field and phase sensitivity but consumes a reference. Stark sensing accesses detuned or low-frequency fields but is quadratic unless biased. The estimand decides which compromise is appropriate.

An atom-referenced transducer need not offer quantum-enhanced precision. Large dipoles, optical spatial selectivity, and reconfigurable carrier coverage can be valuable without outperforming a resource-matched receiver. Such a comparison needs the same aperture, bandwidth, prior information, reference resources, and estimation loss.

Exercises

1. Convert a resolved splitting

A coupling-laser scan resolves a resonant splitting of $2.40\ \mathrm{MHz}$. The effective peak-field dipole is $750\,e a_0$. Find the local peak and RMS fields.

Solution

Using $e a_0=8.47835\times10^{-30}\ \mathrm{C\,m}$,

$$d_{\mathrm{eff}} = 6.3588\times10^{-27}\ \mathrm{C\,m}.$$

Therefore

$$E_0 = \frac{h(2.40\times10^6\ \mathrm{Hz})}{d_{\mathrm{eff}}} = 0.2501\ \mathrm{V\,m^{-1}}.$$

For a linear sinusoid,

$$E_{\mathrm{rms}} = \frac{E_0}{\sqrt2} = 0.1768\ \mathrm{V\,m^{-1}}.$$

2. Correct a detuned doublet

An RF-dressed spectrum gives $S_\omega/(2\pi)=8.0\ \mathrm{MHz}$ while an independent frequency measurement gives $|\Delta_\mu|/(2\pi)=4.8\ \mathrm{MHz}$. Find $|\Omega_\mu|/(2\pi)$.

Solution

The generalized gap obeys

$$S_\omega^2 = \Delta_\mu^2+|\Omega_\mu|^2.$$

The common factor $2\pi$ cancels, so

$$\frac{|\Omega_\mu|}{2\pi} = \sqrt{(8.0\ \mathrm{MHz})^2-(4.8\ \mathrm{MHz})^2} = 6.4\ \mathrm{MHz}.$$

Using $8.0\ \mathrm{MHz}$ directly would overestimate the field by $25\%$.

3. Derive the probe-scan mapping

Take counter-propagating beams with signed wave vectors $k_p>0$ and $k_c<0$. The coupling laser is fixed on resonance. Use the velocity-class condition $\Delta_p+k_pv=0$ to express the two-photon detuning in terms of the scanned probe detuning.

Solution

The selected velocity is

$$v = -\frac{\Delta_p}{k_p}.$$

With $\Delta_c=0$,

$$\begin{aligned} \delta_2(v) &= \Delta_p+(k_p+k_c)v\\ &= \Delta_p - \frac{k_p+k_c}{k_p}\Delta_p\\ &= -\frac{k_c}{k_p}\Delta_p. \end{aligned}$$

Because $-k_c/k_p=|k_c|/k_p=\nu_c/\nu_p$, a two-photon separation $S_\omega$ appears on the probe axis as

$$\Delta\omega_p = \frac{\nu_p}{\nu_c}S_\omega.$$

4. Identify a polarization blind spot

A prepared state has only a $\pi$-allowed RF transition relative to the bias axis. What does an Autler–Townes measurement reveal about a purely transverse field?

Solution

A $\pi$ transition couples to the spherical component $E_0$, parallel to the quantization axis. A purely transverse field has only $q=\pm1$ components, so the chosen transition has $\Omega_\mu=0$ in the ideal selection-rule model. The absence of splitting is a polarization blind spot, not evidence that the total electric field vanishes. Additional $\sigma^+$ and $\sigma^-$ pathways, or a rotated quantization axis, are needed for reconstruction.

5. Fisher information from photon counts

At one operating point the mean detected count is $\mu(E)=\mu_0(1+aE)$ over the local linear range. Assuming Poisson counts, find the single-shot Fisher information and its small-field limit.

Solution

For a Poisson mean,

$$F_E = \frac{1}{\mu(E)} \left( \frac{d\mu}{dE} \right)^2.$$

Since $d\mu/dE=\mu_0a$,

$$F_E = \frac{\mu_0a^2}{1+aE}.$$

At $|aE|\ll1$,

$$F_E \simeq \mu_0a^2, \qquad \sigma_E \geq \frac{1}{|a|\sqrt{\mu_0}}.$$

This bound assumes local unbiasedness and $1+aE>0$. It describes the specified photon likelihood; calibration and technical noise require their own nuisance parameters.

6. Show how a local oscillator restores linear response

Suppose the optical observable is $y(E)=y_0+aE^2$. Set $E(t)=E_L+E_s\cos(\omega_{\mathrm{IF}}t+\phi)$ with $E_s\ll E_L$. Find the component at $\omega_{\mathrm{IF}}$.

Solution

Expanding to first order in $E_s$ gives

$$\begin{aligned} y(t) &= y_0+a \left[ E_L+E_s\cos(\omega_{\mathrm{IF}}t+\phi) \right]^2\\ &\simeq y_0+aE_L^2 +2aE_LE_s \cos(\omega_{\mathrm{IF}}t+\phi). \end{aligned}$$

The intermediate-frequency amplitude is $2|a|E_LE_s$, linear in the weak signal. Its phase is measured relative to the local oscillator.

7. Propagate an external-field uncertainty

A resonant measurement has independent relative standard uncertainties of $0.8\%$ from splitting, $0.4\%$ from the scan factor, $0.6\%$ from the dipole, and $2.5\%$ from the cell transfer. Ignore model discrepancy. Find the combined relative uncertainty.

Solution

Add the independent relative variances:

$$\begin{aligned} \frac{u(E)}{E} &= \sqrt{ (0.008)^2+(0.004)^2 +(0.006)^2+(0.025)^2 }\\ &= 0.0272. \end{aligned}$$

The combined standard uncertainty is about $2.7\%$, dominated by the cell transfer. Improving the spectral fit alone would have little effect.

8. Spatial averaging of a quadratic response

Two equally weighted, noninteracting optical regions experience same-phase fields $E_1=E(1+\epsilon)$ and $E_2=E(1-\epsilon)$. Their weak, additive Stark signals are proportional to $E_i^2$. If a uniform-field model is fitted, what amplitude does it infer? Assume propagation introduces no additional nonlinearity.

Solution

The mean signal is proportional to

$$\frac{E_1^2+E_2^2}{2}=E^2(1+\epsilon^2).$$

The fitted positive amplitude is therefore $E_{\mathrm{fit}}=|E|\sqrt{1+\epsilon^2}$, whereas the mean signed field is $E$. For small inhomogeneity the relative bias is $E_{\mathrm{fit}}/|E|-1=\epsilon^2/2+O(\epsilon^4)$. At $\epsilon=0.5$, the excess is about $11.8\%$. The result measures a weighted RMS field in this particular model. Autler–Townes spectra and finite-depth propagation need their own forward average; the same effective-field rule cannot simply be reused.

References

• S. M. Bohaichuk et al., “Origins of Rydberg-atom electrometer transient response and its impact on radio-frequency pulse sensing,” Physical Review Applied 18, 034030 (2022).

• H. Fan et al., “Effect of vapor-cell geometry on Rydberg-atom-based measurements of radio-frequency electric fields,” Physical Review Applied 4, 044015 (2015).

• C. L. Holloway et al., “Broadband Rydberg atom-based electric-field probe for SI-traceable, self-calibrated measurements,” IEEE Transactions on Antennas and Propagation 62, 6169–6182 (2014).

• Y.-Y. Jau and T. Carter, “Vapor-cell-based atomic electrometry for detection frequencies below 1 kHz,” Physical Review Applied 13, 054034 (2020).

• M. Jing et al., “Atomic superheterodyne receiver based on microwave-dressed Rydberg spectroscopy,” Nature Physics 16, 911–915 (2020).

• S. Kumar, H. Fan, H. Kübler, J. Sheng, and J. P. Shaffer, “Atom-based sensing of weak radio frequency electric fields using homodyne readout,” Scientific Reports 7, 42981 (2017).

• D. H. Meyer, Z. A. Castillo, K. C. Cox, and P. D. Kunz, “Assessment of Rydberg atoms for wideband electric field sensing,” Journal of Physics B 53, 034001 (2020).

• N. Schlossberger et al., “Rydberg states of alkali atoms in atomic vapour as SI-traceable field probes and communications receivers,” Nature Reviews Physics 6, 606–620 (2024).

• N. Schlossberger et al., “Zeeman-resolved Autler–Townes splitting in Rydberg atoms with tunable resonances and a single transition dipole moment,” Physical Review A 109, L021702 (2024).

• J. A. Sedlacek, A. Schwettmann, H. Kübler, R. Löw, T. Pfau, and J. P. Shaffer, “Microwave electrometry with Rydberg atoms in a vapour cell using bright atomic resonances,” Nature Physics 8, 819–824 (2012).

• J. A. Sedlacek, A. Schwettmann, H. Kübler, and J. P. Shaffer, “Atom-based vector microwave electrometry using rubidium Rydberg atoms in a vapor cell,” Physical Review Letters 111, 063001 (2013).

• M. T. Simons, A. H. Haddab, J. A. Gordon, and C. L. Holloway, “A Rydberg atom-based mixer: Measuring the phase of a radio frequency wave,” Applied Physics Letters 114, 114101 (2019).